Tuesday, December 09, 2014

Northern Lights part 1 - Testing the high ISO capability of your camera

At this time of year lots of our clients are off in search of the Northern Lights, and that can be a demanding challenge for both your camera and your lens. It's not like taking photos in daylight, and you're often faced with having to make difficult trade-offs between aperture, shutter speed and ISO. You aperture is of course limited by your lens; you don't want the shutter speed to be too long, so that you can capture the movement of the aurora; but you don't want the ISO to be too high, in case your images are too noisy.

The difficult part of this is sometimes the ISO. How many of us know how high we can push the ISO setting on our cameras before the images start to become degraded? It's not something you can just look up on the internet, for two reasons. Firstly different people have different standards when it comes to the acceptability of noise. And secondly - as we'll see in a minute - it depends on the size at which you're viewing your photos. But fortunately it's not too hard to test for yourself.

The basic idea is that we choose a test scene and photograph it repeatedly using different ISO settings. Ideally the scene should contain a range of colours and a range of dark and light tones. You'll need to set up the camera on a tripod to ensure that each shot is framed the same. And you'll need it to be indoors, because shooting at high ISO values in sunlight is difficult - it often needs a faster shutter speed than your camera can manage.

Something like this, perhaps.

Set the camera to aperture priority, and set the lens to a middling aperture like f/5.6 or f/8 where it's going to perform well. (This test isn't about the lens.) You may as well set the camera to produce JPEGs rather than RAWs because it's quicker and we don't need to do any clever processing. And it doesn't really matter whether you have in-camera noise reduction set to on or off, but remember that your results only apply to that one situation. (You might want to try the test twice, with noise reduction on and off.)

And the procedure is really simple. Set the ISO to 100 and take the shot. Set the ISO to 200 and take the shot. (The exposure will be twice as fast, obviously.) Set the ISO to 400 and take the shot. And so on until you've reached the highest ISO value that your camera can manage.

Now all we need to do is look at the images and see where they start to get noisy. But there's an important point to remember. The closer you look at them (i.e. the higher the magnification), the more you'll see the noise.

Here's an example, using a set of images from a Canon 650D. I've cropped a section from each image from ISO 100 to ISO 6400, and put them together on this composite. This comparison is at 100% resolution - what some people call a 100% crop - because each pixel in this image corresponds to one pixel on the sensor. It looks a bit clumsy having such a big image in this blog post, but it's important to view the composite image at full size (1800 x 900 pixels) if the conclusions are going to be valid.

(Don't worry about the ISO 400 image looking a bit blurry, by the way. It looks like the tripod wasn't perfectly stable, or somebody jogged the camera. Ideally we'd have been a bit more careful. But it doesn't really matter.)

So what do we see here? Well, personally, I think that up to ISO 800 everything is fine. But there's clearly some noise at ISO 1600, and from ISO 3200 upwards it's very bad.

But, remember I said that the size at which you're viewing your photos makes a difference? This test was done at very high resolution - the full 18-megapixel image of the test scene, 5184 x 3456 pixels, would be about 4 feet / 1.3m wide at this resolution if you're reading this on a standard computer monitor. If you were to print it, you wouldn't be able to see all this detail unless you printed at A3 or larger. Most of the time, most of us don't do that.

So what happens if the image is smaller? In simple terms, the noise is reduced. Making an image smaller involves combining several pixels into one pixel, and in that process all the random noise gets averaged out.

Here's another comparison of the images taken at different ISO settings, but this time it's presented at 50% resolution. I've scaled the full test image to 50% of its original size - from 5184x3456 to 2592x1728 pixels - and then extracted the crops.

At this resolution, the full image would be about 27"x17" on a standard resolution monitor, and that isn't massively more than the kinds of monitors some of us use these days with desktop machines. You could theoretically see the individual pixels if you were to print it at 9"x6" or larger. To my eye, averaging out the noise by reducing the size of the image seems to have helped. ISO 6400 is still dreadful and ISO 3200 is still bad, but ISO 1600 isn't looking so bad.

And finally, here's a third comparison which is based on an even greater reduction in image size. This one is at 20% resolution. I've scaled the full test image to 20% of its original size - from 5184x3456 to 1037x691 pixels - and then extracted the crops. So each pixel here is constructed from a 5x5 block of pixels in the original image, and we might expect that the averaging process has done quite a lot to the noise levels.

This is more like the kind of resolution you'd use for displaying the image on the internet. To my eyes, ISO 1600 is quite acceptable at this resolution, and ISO 3200 certainly isn't too bad, though ISO 6400 is still pretty ropey.

So all in all, that gives us quit an interesting set of results:

  • you can use ISO values up to 800 without any concerns;
  • for large prints, ISO 800 is fine but ISO 1600 is questionable;
  • for small prints, or for full-screen display on a large monitor, ISO 1600 is probably acceptable;
  • for use on websites etc ISO 1600 is fine and even ISO 3200 is reasonably OK;
  • but ISO 6400 should be avoided unless you're desperate.
Remember, this is based on MY assessment as to what is or is not acceptable: you may have different standards. But you'd still get a similar pattern with graduations in it depending on the intended use of the images.

And remember also, that this is based on a Canon 650D with no in-camera noise reduction. If you'd like to know how your camera compares ... what's stopping you?


  1. I use a Canon 6D - ISO-12800 for web use it no problem.

    1. Cheers Scott. Have you worked out what ISO setting you could use for prints?

  2. What are your thoughts on the use of noise reduction software in post-processing? I know it removes some detail, but also a lot of noise too...

    1. I think it's entirely subjective, Sophie. Do whatever gives you the results that you like.

      Personally I don't use high-ISO noise reduction in camera. I shoot RAW, so I can always sort out any noise in Lightroom later if it's a problem, and I can decide how much to fix and how much to leave. However I do tend to use long exposure noise reduction in camera, so long as I'm not shooting a star trail or a timelapse or anything like that, because that's harder to do in post-processing and I don't think it has any real downside as far as image quality is concerned.